Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MARs: Multi-view Attention Regularizations for Patch-based Feature Recognition of Space Terrain

Timothy Chase, Karthik Dantu

TL;DR

This work thoroughly analyzes many modern metric learning losses with and without MARs and introduces Multi-view Attention Regularizations (MARs) to constrain the channel and spatial attention across multiple feature views, regularizing the what and where of attention focus.

Abstract

The visual detection and tracking of surface terrain is required for spacecraft to safely land on or navigate within close proximity to celestial objects. Current approaches rely on template matching with pre-gathered patch-based features, which are expensive to obtain and a limiting factor in perceptual capability. While recent literature has focused on in-situ detection methods to enhance navigation and operational autonomy, robust description is still needed. In this work, we explore metric learning as the lightweight feature description mechanism and find that current solutions fail to address inter-class similarity and multi-view observational geometry. We attribute this to the view-unaware attention mechanism and introduce Multi-view Attention Regularizations (MARs) to constrain the channel and spatial attention across multiple feature views, regularizing the what and where of attention focus. We thoroughly analyze many modern metric learning losses with and without MARs and demonstrate improved terrain-feature recognition performance by upwards of 85%. We additionally introduce the Luna-1 dataset, consisting of Moon crater landmarks and reference navigation frames from NASA mission data to support future research in this difficult task. Luna-1 and source code are publicly available at https://droneslab.github.io/mars/.

MARs: Multi-view Attention Regularizations for Patch-based Feature Recognition of Space Terrain

TL;DR

This work thoroughly analyzes many modern metric learning losses with and without MARs and introduces Multi-view Attention Regularizations (MARs) to constrain the channel and spatial attention across multiple feature views, regularizing the what and where of attention focus.

Abstract

The visual detection and tracking of surface terrain is required for spacecraft to safely land on or navigate within close proximity to celestial objects. Current approaches rely on template matching with pre-gathered patch-based features, which are expensive to obtain and a limiting factor in perceptual capability. While recent literature has focused on in-situ detection methods to enhance navigation and operational autonomy, robust description is still needed. In this work, we explore metric learning as the lightweight feature description mechanism and find that current solutions fail to address inter-class similarity and multi-view observational geometry. We attribute this to the view-unaware attention mechanism and introduce Multi-view Attention Regularizations (MARs) to constrain the channel and spatial attention across multiple feature views, regularizing the what and where of attention focus. We thoroughly analyze many modern metric learning losses with and without MARs and demonstrate improved terrain-feature recognition performance by upwards of 85%. We additionally introduce the Luna-1 dataset, consisting of Moon crater landmarks and reference navigation frames from NASA mission data to support future research in this difficult task. Luna-1 and source code are publicly available at https://droneslab.github.io/mars/.
Paper Structure (12 sections, 7 equations, 4 figures, 3 tables)

This paper contains 12 sections, 7 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Learning Framework
  5. Network Architectures
  6. Forming Attention Similarity Constraints
  7. Evaluation
  8. Datasets
  9. Experiments
  10. Implementation Details
  11. Experimental Results
  12. Conclusion

Figures (4)

  • Figure 1: Patch-based features of space terrain exhibit extreme inter-class similarity and varying multi-view observations, which is difficult for metric learning to discern where attention focus is disparate. We propose Multi-view Attention Regularizations (MARs) to alleviate this issue and drive the attention of arbitrary viewpoints together.
  • Figure 2: Framework of the proposed Multi-view Attention Regularizations (MARs). MARs aligns the what (channel) and where (spatial) focus of attention between multiple patch-feature views using distinct metric spaces.
  • Figure 3: Mars (a), Moon (b) landmark examples.
  • Figure 4: Attention visualizations with EigenCAM eigencam on Mars Crater trained with Proxy Anchor $\mathop{\mathrm{\mathcal{L}_{\text{ML}}}}\nolimits$.